CSV search export silently truncates at AUTO_PAGINATION_THRESHOLD #18

When /api/v1/search is called with format=csv and no limit, limit defaults to AUTO_PAGINATION_THRESHOLD (5000, app/core/config.py). The CSV response carries no total count and no next-page link, so a query matching more than 5000 records downloads a file that looks complete but isn't (e.g. ?ec_number=1.1.1.1&format=csv).

Related: get_total_count runs on every CSV request (app/routers/search.py) and its result is discarded, costing an extra COUNT over the joined views.

Options:

• Stream the full result set for CSV.
• Send the total (e.g. X-Total-Count) and pagination Link headers, which would also put the count to use.
• Document the limit and require explicit paging.

Raised in review of #17.